Use obtained eu_counts to configure GPU media pipeline on Gen8/Gen9

If it is not supported, it will fall back to the original config.

Signed-off-by: Zhao Yakui <yakui.zhao@intel.com>
Reviewed-by: Charles Daniel <daniel.charles@intel.com>

6 files changed, 46 insertions, 16 deletions

diff --git a/src/gen8_mfc.c b/src/gen8_mfc.c
index b0ee6fb..d1de92c 100644
--- a/src/gen8_mfc.c
+++ b/src/gen8_mfc.c
@@ -4614,7 +4614,11 @@ Bool gen8_mfc_context_init(VADriverContextP ctx, struct intel_encoder_context *e
     mfc_context->gpe_context.sampler.entry_size = 0;
     mfc_context->gpe_context.sampler.max_entries = 0;
 
-    mfc_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
+    if (i965->intel.eu_total > 0)
+        mfc_context->gpe_context.vfe_state.max_num_threads = 6 * i965->intel.eu_total;
+    else
+        mfc_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
+
     mfc_context->gpe_context.vfe_state.num_urb_entries = 16;
     mfc_context->gpe_context.vfe_state.gpgpu_mode = 0;
     mfc_context->gpe_context.vfe_state.urb_entry_size = 59 - 1;
diff --git a/src/gen8_post_processing.c b/src/gen8_post_processing.c
index 708918b..5ef8cbf 100644
--- a/src/gen8_post_processing.c
+++ b/src/gen8_post_processing.c
@@ -1592,7 +1592,10 @@ gen8_post_processing_context_common_init(VADriverContextP ctx,
     struct pp_module *pp_module;
     struct i965_post_processing_context *pp_context = data;
 
-    pp_context->vfe_gpu_state.max_num_threads = 60;
+    if (i965->intel.eu_total > 0)
+        pp_context->vfe_gpu_state.max_num_threads = 6 * i965->intel.eu_total;
+    else
+        pp_context->vfe_gpu_state.max_num_threads = 60;
     pp_context->vfe_gpu_state.num_urb_entries = 59;
     pp_context->vfe_gpu_state.gpgpu_mode = 0;
     pp_context->vfe_gpu_state.urb_entry_size = 16 - 1;
diff --git a/src/gen8_vme.c b/src/gen8_vme.c
index b14d60a..7a9ed6b 100644
--- a/src/gen8_vme.c
+++ b/src/gen8_vme.c
@@ -1333,6 +1333,7 @@ gen8_vme_context_destroy(void *context)
 
 Bool gen8_vme_context_init(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
 {
+    struct i965_driver_data *i965 = i965_driver_data(ctx);
     struct gen6_vme_context *vme_context = NULL;
     struct i965_kernel *vme_kernel_list = NULL;
     int i965_kernel_num;
@@ -1385,7 +1386,12 @@ Bool gen8_vme_context_init(VADriverContextP ctx, struct intel_encoder_context *e
         vme_context->gpe_context.sampler.entry_size = 0;
         vme_context->gpe_context.sampler.max_entries = 0;
 
-        vme_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
+        if (i965->intel.eu_total > 0) {
+            vme_context->gpe_context.vfe_state.max_num_threads = 6 *
+                               i965->intel.eu_total;
+        } else
+            vme_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
+
         vme_context->gpe_context.vfe_state.num_urb_entries = 64;
         vme_context->gpe_context.vfe_state.gpgpu_mode = 0;
         vme_context->gpe_context.vfe_state.urb_entry_size = 16;
diff --git a/src/gen9_post_processing.c b/src/gen9_post_processing.c
index e9b8f86..1e2d33a 100644
--- a/src/gen9_post_processing.c
+++ b/src/gen9_post_processing.c
@@ -547,10 +547,14 @@ gen9_post_processing_context_init(VADriverContextP ctx,
     gpe_context->surface_state_binding_table.surface_state_offset = ALIGN(MAX_SCALING_SURFACES * 4, 64);
     gpe_context->surface_state_binding_table.length = ALIGN(MAX_SCALING_SURFACES * 4, 64) + ALIGN(MAX_SCALING_SURFACES * SURFACE_STATE_PADDED_SIZE_GEN9, 64);
 
-    if (i965->intel.has_bsd2)
-        gpe_context->vfe_state.max_num_threads = 300;
-    else
-        gpe_context->vfe_state.max_num_threads = 60;
+    if (i965->intel.eu_total > 0) {
+        gpe_context->vfe_state.max_num_threads = i965->intel.eu_total * 6;
+    } else {
+        if (i965->intel.has_bsd2)
+            gpe_context->vfe_state.max_num_threads = 300;
+        else
+            gpe_context->vfe_state.max_num_threads = 60;
+    }
 
     gpe_context->vfe_state.curbe_allocation_size = 37;
     gpe_context->vfe_state.urb_entry_size = 16;
diff --git a/src/gen9_vme.c b/src/gen9_vme.c
index fab80ce..11602a8 100644
--- a/src/gen9_vme.c
+++ b/src/gen9_vme.c
@@ -1978,6 +1978,7 @@ gen9_vme_context_destroy(void *context)
 
 Bool gen9_vme_context_init(VADriverContextP ctx, struct intel_encoder_context *encoder_context)
 {
+    struct i965_driver_data *i965 = i965_driver_data(ctx);
     struct gen6_vme_context *vme_context;
     struct i965_kernel *vme_kernel_list = NULL;
     int i965_kernel_num;
@@ -2037,7 +2038,12 @@ Bool gen9_vme_context_init(VADriverContextP ctx, struct intel_encoder_context *e
     vme_context->gpe_context.sampler.entry_size = 0;
     vme_context->gpe_context.sampler.max_entries = 0;
 
-    vme_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
+    if (i965->intel.eu_total > 0) {
+        vme_context->gpe_context.vfe_state.max_num_threads = 6 *
+                              i965->intel.eu_total;
+    } else
+        vme_context->gpe_context.vfe_state.max_num_threads = 60 - 1;
+
     vme_context->gpe_context.vfe_state.num_urb_entries = 64;
     vme_context->gpe_context.vfe_state.gpgpu_mode = 0;
     vme_context->gpe_context.vfe_state.urb_entry_size = 16;
diff --git a/src/gen9_vp9_encoder.c b/src/gen9_vp9_encoder.c
index 74d0d2f..537b390 100644
--- a/src/gen9_vp9_encoder.c
+++ b/src/gen9_vp9_encoder.c
@@ -3679,9 +3679,12 @@ gen9_vp9_mbenc_kernel(VADriverContextP ctx,
 }
 
 static void
-gen9_init_gpe_context_vp9(struct i965_gpe_context *gpe_context,
+gen9_init_gpe_context_vp9(VADriverContextP ctx,
+                          struct i965_gpe_context *gpe_context,
                           struct vp9_encoder_kernel_parameter *kernel_param)
 {
+    struct i965_driver_data *i965 = i965_driver_data(ctx);
+
     gpe_context->curbe.length = kernel_param->curbe_size; // in bytes
 
     gpe_context->sampler.entry_size = 0;
@@ -3700,7 +3703,11 @@ gen9_init_gpe_context_vp9(struct i965_gpe_context *gpe_context,
     gpe_context->surface_state_binding_table.surface_state_offset = ALIGN(MAX_VP9_ENCODER_SURFACES * 4, 64);
     gpe_context->surface_state_binding_table.length = ALIGN(MAX_VP9_ENCODER_SURFACES * 4, 64) + ALIGN(MAX_VP9_ENCODER_SURFACES * SURFACE_STATE_PADDED_SIZE_GEN9, 64);
 
-    gpe_context->vfe_state.max_num_threads = 112; // 16 EU * 7 threads
+    if (i965->intel.eu_total > 0)
+        gpe_context->vfe_state.max_num_threads = 6 * i965->intel.eu_total;
+    else
+        gpe_context->vfe_state.max_num_threads = 112; // 16 EU * 7 threads
+
     gpe_context->vfe_state.curbe_allocation_size = MAX(1, ALIGN(gpe_context->curbe.length, 32) >> 5); // in registers
     gpe_context->vfe_state.urb_entry_size = MAX(1, ALIGN(kernel_param->inline_data_size, 32) >> 5); // in registers
     gpe_context->vfe_state.num_urb_entries = (MAX_URB_SIZE -
@@ -4606,7 +4613,7 @@ gen9_vme_scaling_context_init_vp9(VADriverContextP ctx,
     scoreboard_param.walkpat_flag = 0;
 
     gpe_context = &scaling_context->gpe_contexts[0];
-    gen9_init_gpe_context_vp9(gpe_context, &kernel_param);
+    gen9_init_gpe_context_vp9(ctx, gpe_context, &kernel_param);
     gen9_init_vfe_scoreboard_vp9(gpe_context, &scoreboard_param);
 
     scaling_context->scaling_4x_bti.scaling_frame_src_y = VP9_BTI_SCALING_FRAME_SRC_Y;
@@ -4632,7 +4639,7 @@ gen9_vme_scaling_context_init_vp9(VADriverContextP ctx,
     kernel_param.sampler_size = 0;
 
     gpe_context = &scaling_context->gpe_contexts[1];
-    gen9_init_gpe_context_vp9(gpe_context, &kernel_param);
+    gen9_init_gpe_context_vp9(ctx, gpe_context, &kernel_param);
     gen9_init_vfe_scoreboard_vp9(gpe_context, &scoreboard_param);
 
     memset(&scale_kernel, 0, sizeof(scale_kernel));
@@ -4674,7 +4681,7 @@ gen9_vme_me_context_init_vp9(VADriverContextP ctx,
     scoreboard_param.walkpat_flag = 0;
 
     gpe_context = &me_context->gpe_context;
-    gen9_init_gpe_context_vp9(gpe_context, &kernel_param);
+    gen9_init_gpe_context_vp9(ctx, gpe_context, &kernel_param);
     gen9_init_vfe_scoreboard_vp9(gpe_context, &scoreboard_param);
 
     memset(&scale_kernel, 0, sizeof(scale_kernel));
@@ -4722,7 +4729,7 @@ gen9_vme_mbenc_context_init_vp9(VADriverContextP ctx,
         } else
             scoreboard_param.walkpat_flag = 0;
 
-        gen9_init_gpe_context_vp9(gpe_context, &kernel_param);
+        gen9_init_gpe_context_vp9(ctx, gpe_context, &kernel_param);
         gen9_init_vfe_scoreboard_vp9(gpe_context, &scoreboard_param);
 
         memset(&scale_kernel, 0, sizeof(scale_kernel));
@@ -4762,7 +4769,7 @@ gen9_vme_brc_context_init_vp9(VADriverContextP ctx,
 
     for (i = 0; i < NUM_VP9_BRC; i++) {
         gpe_context = &brc_context->gpe_contexts[i];
-        gen9_init_gpe_context_vp9(gpe_context, &kernel_param);
+        gen9_init_gpe_context_vp9(ctx, gpe_context, &kernel_param);
         gen9_init_vfe_scoreboard_vp9(gpe_context, &scoreboard_param);
 
         memset(&scale_kernel, 0, sizeof(scale_kernel));
@@ -4801,7 +4808,7 @@ gen9_vme_dys_context_init_vp9(VADriverContextP ctx,
     scoreboard_param.walkpat_flag = 0;
 
     gpe_context = &dys_context->gpe_context;
-    gen9_init_gpe_context_vp9(gpe_context, &kernel_param);
+    gen9_init_gpe_context_vp9(ctx, gpe_context, &kernel_param);
     gen9_init_vfe_scoreboard_vp9(gpe_context, &scoreboard_param);
 
     memset(&scale_kernel, 0, sizeof(scale_kernel));